Wanda Li PhD Defence

PhD Thesis Abstract: From Reactive Design-Based Taxonomies to Predictive Metrics-Driven Deceptive Pattern Frameworks.

When navigating digital services, users can encounter designs that can alter their
decisions in subtle ways that are difficult to detect [125]. Organizations employ these
designs to serve their own interests, sometimes at the expense of users [28, 101, 125,
189]. These designs are called Deceptive Patterns.
To understand Deceptive Patterns, experts developed taxonomies [28]. Since
their introduction, taxonomies have provided the foundation for identifying Deceptive
Patterns, enabling experts to evaluate their prevalence, create mitigation tools, and
hold organizations accountable [28, 72, 125].
However, existing taxonomies do not enable consistent identification, a core function
of classification systems [16, 162]. As existing taxonomies rely on textual descriptions
of designs (design-based), they struggle to capture visual nuance and variation
across platforms and domains. This creates interpretive ambiguity, causing experts
to misidentify benign designs or overlook harmful ones, which obscures the harms
of Deceptive Patterns. The reactive nature of design-based taxonomies compounds
these issues, as experts can only update them after finding new designs [125]. Without
tools that anticipate new designs, these taxonomies cannot keep pace with evolving
interfaces or the continual emergence of new Deceptive Patterns, reducing their longterm
utility for identification [28]. These identification inconsistencies can undermine
the credibility of findings and weaken comparisons across studies.
This dissertation addresses this problem by proposing methodological improvements
that enhance identification consistency when applying existing design-based
taxonomies. Starting with existing taxonomies remains essential because they are the
most widely adopted tools for identifying Deceptive Patterns [70]. A large-scale evaluation
of 143 mobile apps tested these improvements, revealing both benefits and persistent
limitations. These findings motivated the development of the Metrics-Driven
Deceptive Patterns Framework, which grounds identification in the Digital Metrics
and KPIs that Deceptive Patterns aim to optimize [28, 138, 188]. The framework
shifts focus from designs to organizational intent, establishing a structured relationship:
Digital Metrics and KPIs → Deceptive Patterns → User Outcomes. Validation
across multiple datasets shows that the framework improves identification consistency,
clarifies harms, and predicts locations of Deceptive Patterns based on organizational
incentives. Together, these findings demonstrate that the Metrics-Driven Deceptive
Patterns Framework not only overcomes key limitations of existing design-based
taxonomies that are barriers to consistent identification but also provides enhanced
explanatory and predictive power for research and policy.